search

tum-vision / fastfusion

Volumetric 3D Mapping in Real-Time on a CPU
GNU General Public License v2.0
559 stars 173 forks source link

How to deal with outliers? #1

Open ghost opened 10 years ago

ghost commented 10 years ago

Hi Frank and others,

the "fastfusion" project is awesome! Thank you very much for sharing. How do you deal with outliers, e.g. sparse point measurements away from an object? I am looking for a solution to be used with stereo cameras, which are not as precise as the pattern systems aka. Kinect.

Is there a process reducing the "occupancy probability" of a node (=Brick, or Voxel) if no measurements are put into it after some time?

Thanks again!

sturmju commented 10 years ago

Hi,

I think that there is already a simple outlier supression that ignores voxels with very few observations. Frank is currently at ICRA but maybe he can point you to the corresponding line in the source code..

Cheers Jürgen

On Mon, Jun 2, 2014 at 8:45 PM, mojovski notifications@github.com wrote:

Hi Frank and others,

the "fastfusion" project is awesome! Thank you very much for sharing. How do you deal with outliers, e.g. sparse point measurements away from an object? I am looking for a solution to be used with stereo cameras, which are not as precise as the pattern systems aka. Kinect.

Is there a process reducing the "occupancy probability" of a node (=Brick, or Voxel) if no measurements are put into it after some time?

Thanks again!

— Reply to this email directly or view it on GitHub https://github.com/tum-vision/fastfusion/issues/1.